Non-Invasive Machine Health Diagnosis

Abstract

In industrial applications, the machines vital for smooth operations require continuing maintenance and repair to maintain acceptable performance. Therefore, the present work is a research effort to investigate the effectiveness of vibration technique in a non-intrusive inspection (NII) manner when monitoring complex machine scenarios with the need for accuracy and resolution limitation in mind. The diagnostic effectiveness was investigated in comparison with the most commonly used conventional invasive sensing technique. A careful selection was made after a comprehensive literature review of existing NII techniques and their potential for health diagnosis. The complex machine diagnostics scenarios were also selected after a detailed review of existing literature, highlighting their relevance in both an academic and an industrial context. The review identified the basis for a test environment that not only demonstrated the effectiveness of NII but also introduced a novel diagnostic research problem into the selected machine fault scenario. In this research, we chose a radar sensing technique (RST) due to it non-invasive and other options it has been selected to be the NII technique candidate to measure the vibration response of a gear test rig in parallel with a contact-based accelerometer. The latter is the mostly commonly used invasive sensing element for rotatory machinery in industry. Gears are a significant component for any mechanical machine and in this research were tested under different lubrications, speeds, loads, seeded tooth faults and misalignments. These tests were able to demonstrate the effectiveness of the NII technique in a complex machine diagnostic environment. An extensive experimental scheme of 144 experiments with different speeds, values of load, viscosities of lubricant, degree of misalignment and levels of the seeded fault. The measurements by radar sensor and accelerometer were latter analysed and assessed for their diagnostic potential. Novel unified models used in the lab during testing and analysing from the measured data of both the sensing approaches were developed using regression and superposition. These models can predict the variations in gear vibration spectra if a quantitative change in lubrication viscosity, torque load, speed and misalignment is known. The acquired data from both invasive and non-invasive sensing methods were compared using regression fitting, Root Mean Square (RMS) and goodness of fit measurement with R-square (R2). The findings reveal significant agreement between the two methods. For the different speed tests, the goodness of fit for accelerometer and RST in terms of R2 values were 0.99 and 0.96 respectively. Similarly values of R2 were found for the different load, misalignment, lubrication, and seeded fault tests which enabled a critical evaluation of the relative effectiveness and competitiveness of both the invasive and non-invasive sensing elements. The polynomials obtained from data regression were validated on arbitrary test scenarios and respective accuracies of both sensing elements are discussed. The relative prediction accuracy of the models was observed to vary significantly. In the case of an arbitrary speed value, the accelerometer invasive sensing model exhibited 85% accuracy, while RST non-invasive sensing model showed an accuracy of 71%. In contrast, in the case of arbitrary load values, the accuracy of the RST model was 3% higher than the accelerometer model. The observations when validating the effectiveness of both models under different misalignment conditions are also discussed. The RST is proposed as a promising technique for detecting and diagnosing machine health. However, due to the wide range of objects that radar can detect simultaneously, the quality of the acquired signal still needs improvement. Enhanced filtration methods are necessary to eliminate noise and improve signal clarity.